A cross-border e-commerce consumption voucher generation method, device, medium and product

Abstract

本说明书提供了一种跨境电商消费券的生成方法、设备、介质及产品。通过利用基于历史多维数据训练的三个预测模型来分别获取消费券的预测领取率、核销率与回报率，每个预测模型能够基于已学习到的对应服务环节中的复杂规律，实现对动态数据的准确预测。基于上述预测值确定各消费券的预期引导成交金额，并以最大化全局成交总金额为目标自动从候选集合中确定目标消费券集合，从而将复杂的规则与约束转化为可求解的优化问题，以进行快速寻优。通过将选定的消费券转换成符合目标币种的门槛与面额，确保了运营策略在全球多币种环境下的一致性与可用性。依此，本说明书提升了在跨境电商复杂场景下的决策效率、策略质量与运营自动化水平。

Description

Technical Field

[0001] This specification relates to one or more embodiments in the field of data processing technology, and in particular to a method, device, medium and product for generating cross-border e-commerce consumer vouchers. Background Technology

[0002] Currently, the technology for generating cross-border e-commerce vouchers primarily relies on post-hoc statistical analysis of past operational data and the subjective experience of operational personnel. Operations teams typically collect and statistically analyze historical data on the effectiveness of various voucher distributions. Then, based on the current operational plan, referencing the trends presented in the statistical analysis, and combining the operations team's understanding and experience in specific markets, a combination of voucher threshold amounts and discount tiers is designed for different countries. This decision-making mechanism heavily depends on the experience and judgment of operational personnel, making it difficult to integrate and process massive, multi-dimensional, and continuously changing market feedback data. When dealing with cross-border e-commerce operational scenarios involving multiple currencies, multiple markets, and diverse needs, it suffers from shortcomings in decision-making efficiency, strategy accuracy, and process automation. Summary of the Invention

[0003] In view of the above, one or more embodiments of this specification provide the following technical solutions: According to a first aspect of one or more embodiments of this specification, a method for generating cross-border e-commerce consumer vouchers is proposed, comprising: Based on the selected base currency, a candidate set of consumption vouchers corresponding to the base currency is determined, and the consumption vouchers have corresponding base threshold amount and base face value; Based on the selected target countries, for each voucher in the candidate set, the predicted redemption rate, predicted redemption rate, and predicted return rate of the voucher are obtained using pre-trained redemption rate prediction models, redemption rate prediction models, and return rate prediction models, respectively. The redemption rate prediction model, the redemption rate prediction model, or the return rate prediction model is trained using multiple samples with ground truth labels. The ground truth labels of the samples in the redemption rate prediction model are historical redemption rates, the ground truth labels of the samples in the redemption rate prediction model are historical redemption rates, and the ground truth labels of the samples in the return rate prediction model are historical return rates. Based on the predicted redemption rate, predicted redemption rate, and predicted return rate of each of the aforementioned consumption vouchers, the expected transaction amount guided by each of the aforementioned consumption vouchers is determined. Based on the expected transaction amount of each of the aforementioned consumption vouchers, with the goal of maximizing the total global transaction amount, a target consumption voucher set for the target country is determined from the candidate set, wherein the target consumption voucher set includes at least one consumption voucher selected from the candidate set, and the total global transaction amount is the sum of the expected transaction amounts of each of the selected consumption vouchers. Convert the benchmark threshold amount and the benchmark denomination corresponding to each of the selected consumer vouchers into a threshold amount and a denomination in the target currency corresponding to the target country.

[0004] According to a second aspect of one or more embodiments of this specification, an electronic device is provided, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor runs the executable instructions to implement the steps of the method described in the first aspect.

[0005] According to a third aspect of one or more embodiments of this specification, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0006] According to a fourth aspect of one or more embodiments of this specification, a computer program product is provided, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the method described in the first aspect are implemented. As can be seen from the above embodiments, this specification obtains the predicted redemption rate, redemption rate and return rate of consumer vouchers by using three prediction models trained based on historical multi-dimensional data respectively. Each prediction model can accurately predict dynamic data based on the complex rules learned in the corresponding service link. Based on the above predicted values, the expected guided transaction amount of each consumer voucher is determined, and the target consumer voucher set is automatically determined from the candidate set with the goal of maximizing the total global transaction amount, so as to transform complex rules and constraints into an optimizable problem that can be solved for rapid optimization. By converting the selected consumer vouchers in the unified benchmark currency into thresholds and denominations that conform to the target currency, the consistency and usability of the operation strategy in the global multi-currency environment are ensured. Accordingly, this specification improves the decision-making efficiency, strategy quality and operation automation level in the complex scenario of cross-border e-commerce. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a schematic diagram of the architecture of a generation service system for cross-border e-commerce consumer vouchers provided by an exemplary embodiment.

[0008] Figure 2 is a schematic diagram of a method for generating cross-border e-commerce consumer vouchers provided by an exemplary embodiment.

[0009] Figure 3 is a schematic diagram of the training principle of a prediction model provided by an exemplary embodiment.

[0010] Figure 4 is a schematic diagram of the structure of an encoder of a prediction model provided by an exemplary embodiment.

[0011] Figure 5This is a schematic diagram of the structure of the decoder of a prediction model provided in an exemplary embodiment.

[0012] Figure 6 This is a schematic diagram illustrating an application example of a method for generating cross-border e-commerce consumer vouchers provided in an exemplary embodiment.

[0013] Figure 7 This is a schematic structural diagram of a device provided in an exemplary embodiment.

[0014] Figure 8 This is a schematic structural diagram of a cross-border e-commerce consumer voucher generation system provided in an exemplary embodiment.

[0015] Figure 9 This is a schematic structural diagram of a cross-border e-commerce consumer voucher generation device provided in an exemplary embodiment. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0017] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of related data shall comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals shall be provided for users to choose to authorize or refuse.

[0018] The cross-border e-commerce industry needs to provide services to consumers in multiple countries and regions around the world. Consumers in different countries and regions typically use different currencies for transactions, and their consumption habits and spending power also vary significantly. In the global operation scenario of cross-border e-commerce platforms, vouchers are widely used to stimulate consumer demand, increase user activity, and drive the overall transaction volume growth of the platform.

[0019] The technology for generating cross-border e-commerce vouchers primarily relies on post-hoc statistical analysis of past operational data and the subjective experience of operational personnel. The operations team typically collects and statistically analyzes historical data on the effectiveness of various voucher distributions. Then, based on the current operational plan, referencing the trends presented in the statistical analysis, and combining the operations team's understanding and experience in specific markets, a combination of voucher threshold amounts and discount tiers is designed for different countries.

[0020] This method, based on human experience and historical statistics, can be effective when the service scope is limited or the market environment is relatively stable. However, as the cross-border e-commerce market continues to expand and become more complex, currency exchange rates, local consumption levels, and user price sensitivity vary across different countries. Designing and optimizing multi-tiered vouchers for each country using manual statistics and empirical rules is inefficient and difficult to scale. Secondly, this method is a retrospective, static strategy formulation approach. It cannot accurately quantify and balance the estimated operating costs of vouchers with the resulting increase in transaction volume during the strategy generation stage, making it difficult to maximize platform revenue under strict budget constraints. Furthermore, when applying a voucher combination designed in one base currency to other countries using different currencies, simple currency conversion often fails to simultaneously maintain consistent discount levels and align with local user preferences. For example, the converted denominations or thresholds may contain non-integer or decimals that do not conform to local spending habits, impacting user experience and promotional effectiveness.

[0021] Therefore, the decision-making mechanism of existing solutions relies heavily on the experience and judgment of operations personnel, making it difficult to integrate and process massive, multi-dimensional, and continuously changing market feedback data. Furthermore, they lack the ability to quickly model and optimize various rules and resource constraints during the strategy generation phase, failing to automatically and efficiently determine the optimal or near-optimal combination of consumer vouchers under multiple objectives such as cost, revenue, and localization preferences. When dealing with cross-border e-commerce operation scenarios involving multiple currencies, multiple markets, and diverse needs, they are inadequate in terms of decision-making efficiency, strategy accuracy, and process automation.

[0022] Figure 1 This is a schematic diagram of the architecture of a cross-border e-commerce consumer voucher generation service system provided in an exemplary embodiment. For example... Figure 1 As shown, the system may include a server 11, a network 12, and several electronic devices, such as a personal computer (PC) 13, a mobile phone 14, etc.

[0023] Server 11 can be a physical server containing an independent host, or it can be a virtual server hosted in a host cluster. During operation, server 11 can run server-side programs for a certain application to implement the relevant functions of that application. For example, when server 11 runs a program for generating cross-border e-commerce vouchers, it can become a corresponding cross-border e-commerce voucher generation service platform.

[0024] PC13 and mobile phone14 are just some of the types of electronic devices that users can use. In reality, users can obviously also use electronic devices such as tablets, laptops, PDAs (Personal Digital Assistants), wearable devices (such as smart glasses, smartwatches, etc.), etc., and one or more embodiments in this specification do not limit this. During operation, the electronic device can run a client-side program of an application to achieve the relevant functions of that application. For example, when the electronic device runs a program for generating cross-border e-commerce vouchers, it can act as a client for that cross-border e-commerce voucher generation service. The aforementioned client application for the cross-border e-commerce voucher generation service can be launched and run on the electronic device. This client-side program can be a native application installed on the electronic device, or it can be a mini-program, quick app, or other similar form. Of course, when using web technologies such as HTML5 or similar, the relevant functions can be achieved through a page displayed by a browser. This browser can be a standalone browser application or a browser module embedded in some applications.

[0025] As for the network 12 that enables interaction between electronic devices such as PC13 and mobile phone 14 and server 11, communication can be achieved using either wired or wireless networks, depending on the communication methods supported by the respective electronic devices. This specification does not impose any restrictions on this. For example, PC13 can support both wired and wireless communication, so it can use either wired or wireless networks as needed. Mobile phone 14 typically only supports wireless communication, so it can use a wireless network for communication.

[0026] Figure 2 This is a schematic diagram illustrating a method for generating cross-border e-commerce consumer vouchers, provided as an exemplary embodiment. Figure 2 As shown, the generation method includes the following steps: S200: Based on the selected base currency, determine the candidate set of consumption vouchers corresponding to the base currency.

[0027] Among them, the consumption vouchers have corresponding benchmark threshold amounts and benchmark denominations, which are set for benchmark currencies; The candidate set is a collection of vouchers containing various combinations of threshold amounts and face values. Each voucher in the candidate set is uniquely determined by a combination of a baseline threshold amount and a baseline face value denominated in a base currency. Therefore, for any two vouchers, if at least one of their baseline threshold amount or baseline face value is different, they are considered different vouchers and exist as independent elements in the candidate set.

[0028] The consumption voucher values ​​in the candidate set of this specification are all denominated in a base currency, enabling unified budget control at the platform level and comparative analysis of the effects of consumption vouchers across borders. For example, the base currency can be major currencies such as the US dollar, the Chinese yuan, or the euro. In contrast, if each country independently uses its own local currency candidate set for optimization, the subsidy costs and expected returns of consumption voucher programs in each country will be on different currency scales, making it more difficult to allocate budgets at the platform level and assess the ROI of overall global operations.

[0029] After determining the candidate set of consumption vouchers under the benchmark currency, each candidate in the set needs to be quantitatively evaluated. The purpose of the evaluation is to calculate the expected transaction volume driven by each consumption voucher. The expected transaction volume driven refers to the total additional merchandise volume (GMV) that the platform is expected to generate in a specific target country market after issuing the consumption voucher and deducting the corresponding subsidy costs (i.e., the benchmark face value). This serves as a basis for measuring the investment efficiency of each consumption voucher.

[0030] In some implementations, a fitting function can be constructed based on historical data, derived from the characteristics of consumption vouchers and market features, to the final transaction amount. This approach relies on predefined or mined mathematical relationships between features and target variables. For example, statistical analysis can be used to determine the correlation between variables such as the threshold amount, face value, and market attributes of the issuing country of consumption vouchers and historical transaction amounts, thereby constructing a function model. The parameters of this model are obtained by fitting historical data and used to calculate the expected transaction amount of new consumption vouchers in the target country. The focus of this approach is on feature engineering and the assumptions made in the model form; its predictive performance largely depends on how closely the selected pre-defined functional relationship approximates actual patterns.

[0031] In other implementations, the actual transaction amounts generated after the issuance of historical consumption vouchers can be collected as labels to train an end-to-end predictive model. This type of predictive model takes the feature combinations of consumption vouchers and corresponding market environment indicators as input, and through learning a nonlinear mapping, outputs an estimate of the expected transaction amount generated. Compared to the aforementioned fitting function approach, this method can automatically learn feature interactions and higher-order patterns from the data, without relying heavily on manually preset function model structures.

[0032] However, whether constructing a parametric fitting function or employing an end-to-end prediction model, both methods treat the entire process from voucher distribution to final transaction as a single entity, and the method of constructing the fitting function relies heavily on discrete human experience. In actual service delivery, the process from distribution to final transaction involves different service stages: user redemption, decision-making for redemption, and the calculation of the transaction amount after redemption. These methods couple data from multiple service stages into a single model for unified prediction, making it difficult to finely distinguish and accurately characterize the unique patterns of each stage. Especially in scenarios with imbalanced samples, this can easily affect the accuracy and stability of the final prediction results.

[0033] The method used in this manual is shown in steps S202 and S204 below, which independently predicts the redemption rate, redemption rate and return rate of the consumption vouchers, and then calculates the expected transaction amount they will guide.

[0034] S202: Based on the selected target countries, for each voucher in the candidate set, use pre-trained redemption rate prediction model, redemption rate prediction model, and return rate prediction model to obtain the predicted redemption rate, predicted redemption rate, and predicted return rate of the voucher, respectively.

[0035] The redemption rate prediction model, redemption rate prediction model, and return on investment prediction model are trained using multiple samples with ground truth labels. Each sample includes the threshold amount, face value, and country of distribution of a historical consumption voucher, as well as the distribution indicators obtained at a specified time point. The distribution indicators are metrics that characterize the distribution status of historical consumption vouchers at the specified time point. The ground truth label for the samples in the redemption rate prediction model is the historical redemption rate, the ground truth label for the samples in the redemption rate prediction model is the historical redemption rate, and the ground truth label for the samples in the return on investment prediction model is the historical return on investment. The return on investment (ROI) is calculated by dividing the actual transaction amount of users using consumption vouchers by the face value of the consumption vouchers, and is used to evaluate the efficiency of consumption vouchers in stimulating user transactions.

[0036] This specification utilizes three pre-trained prediction models to predict the redemption rate, redemption rate, and return on investment, respectively. These three prediction models will be referred to below as the redemption rate prediction model, the redemption rate prediction model, and the return on investment prediction model. Regarding the model architecture, the three prediction models can use the same architecture or different architectures. For example, the prediction models can be implemented using machine learning model architectures such as linear regression, regression tree models, or deep neural networks.

[0037] Taking three prediction models all employing a neural network architecture as an example, each prediction model is a supervised learning model that gains predictive ability by learning patterns from a large amount of historical data. Model training is based on samples with ground truth labels. Each sample corresponds to a historical instance of voucher distribution, containing both feature parts and ground truth labels. The feature parts describe the key information of that distribution, while the ground truth labels correspond to the final actual result achieved.

[0038] The sample's features consist of two types of information. The first type is the inherent static attributes of the historical vouchers themselves, specifically including their set threshold amount, face value, and the countries in which they are distributed. These attributes remain unchanged throughout the voucher distribution process, defining the vouchers' basic characteristics and the distribution market. The second type is the distribution metrics obtained at specific points in time during this distribution process. These metrics characterize the distribution status of the vouchers at those designated points in time. For example, distribution metrics include data reflecting the distribution progress (e.g., the number of vouchers issued), the redemption quantity and redemption rate (e.g., reflecting user response), and the distribution duration. Unlike static attributes, the values ​​of distribution metrics change over time, thus reflecting the dynamic changes in the voucher distribution effect.

[0039] The ground truth labels for the samples are set differently depending on the model's prediction objective. For samples used to train the redemption rate prediction model, the ground truth label is the historical redemption rate of the historical voucher. For samples used to train the redemption rate prediction model, the ground truth label is the historical redemption rate. For samples used to train the return on investment prediction model, the ground truth label is the historical return on investment. Through training with a large number of such samples, the model learns the mapping relationship between feature combinations containing static attributes and dynamic process indicators and the corresponding conversion rate or return on investment.

[0040] After the predictive model is trained, when it is necessary to evaluate the value of consumption vouchers from a candidate set in a currently selected target country, a prediction request conforming to the input format of the predictive model can be constructed based on the voucher's baseline threshold amount and baseline face value, combined with pre-set distribution indicators for that target country. This prediction request is then input into three pre-trained models. The redemption rate prediction model outputs the predicted redemption rate for the consumption voucher, the redemption rate prediction model outputs the predicted redemption rate, and the return rate prediction model outputs the predicted return rate. This enables a multi-dimensional quantitative evaluation of the consumption voucher's market conversion effect in the target country.

[0041] This manual uses three predictive models to forecast the redemption rate, redemption rate, and return on investment, allowing each model to focus more intently on learning the complex patterns specific to its corresponding service stage. For example, the redemption rate may be more sensitive to the face value and distribution strategy, the redemption rate is more affected by the threshold amount and the duration of the campaign, while the return on investment reflects the relationship between product mix and user spending power more deeply. Compared to directly predicting the expected transaction amount, this manual achieves higher predictive accuracy through independent and refined modeling of each stage.

[0042] This manual uses three predictive models to predict the redemption rate, redemption rate, and return rate, allowing each model to focus more intently on learning the complex patterns unique to each service stage of the corresponding voucher. These complex patterns stem from the differences in conversion factors and data samples at each service stage. Regarding conversion factors, voucher redemption behavior is primarily driven by factors such as face value, distribution channels, and initial user interest; redemption behavior relies more heavily on factors such as the minimum spending threshold, product price, and the real-time promotional environment; while the return rate, in addition to the above factors, reflects deeper aspects such as product gross profit margin, user spending power, and cross-category purchasing tendency. Regarding data samples, the available historical data for each stage varies in scale, positive / negative sample ratio, and feature dimensions. For example, the redemption behavior data sample size is the largest but also the noisiest, the redemption data sample size is moderate, and the return rate data sample size is relatively small. Coupled with these three vastly different types of data to a single predictive model, feature conflicts and sample bias can easily lead to model convergence difficulties or inaccurate predictions. Therefore, by decoupling the service links and establishing three independent prediction models, this solution can ensure the accuracy and stability of predictions in each link, thereby laying a reliable data foundation for subsequent global optimization.

[0043] S204: Based on the predicted redemption rate, predicted redemption rate, and predicted return rate of each consumer voucher, determine the expected transaction amount to be guided by each consumer voucher.

[0044] For a single voucher, the expected transaction amount it will drive is specifically the product of its base threshold amount, predicted redemption rate, predicted redemption rate, and predicted return rate. Specifically, the predicted redemption rate estimates the expected number of users who will claim the voucher, the predicted redemption rate further estimates the proportion of these users who will ultimately redeem the voucher, and for each redemption, the expected transaction amount can be represented by the product of the voucher's base threshold amount and the predicted return rate. This product reflects the additional transaction amount generated by the voucher's face value in a typical order redeeming the voucher. Therefore, the expected transaction amount driven by a single voucher can be understood as the product of the expected number of users redeeming (determined by the base number of claiming users and the redemption rate) and the expected transaction amount driven by a single redemption (determined by the threshold amount and the return rate).

[0045] S206: Based on the expected transaction amount of each consumption voucher, with the goal of maximizing the total global transaction amount, determine the target consumption voucher set for the selected target countries from the candidate set.

[0046] The target set of consumer vouchers includes at least one consumer voucher selected from the candidate set, and the total transaction amount is the sum of the expected transaction amounts of each selected consumer voucher.

[0047] Based on the expected transaction amount of each voucher in the candidate set determined in step S204, representing the individual value of that voucher, a screening decision can be made to determine the target voucher set. The goal of this decision is to maximize the total global transaction amount, that is, to maximize the sum of the expected transaction amounts of all vouchers ultimately selected into the target set. Therefore, the decision-making process is not simply about selecting the single voucher with the highest value, but rather about finding an "optimal combination" of possible vouchers as the target voucher set from a global perspective of the candidate set, so that the overall contribution (total global transaction amount) of the target voucher set is the highest among all possible combinations.

[0048] S208: Convert the base threshold amount and base denomination of each selected consumption voucher into the threshold amount and denomination in the target currency for the target country.

[0049] After determining the target set of consumption vouchers denominated in a base currency, currency localization is performed to adapt to the local market of the target country. Therefore, it is necessary to convert the base threshold amount and base face value of each selected consumption voucher from a uniform base currency value to an amount expressed in the target currency circulating in the target country.

[0050] As demonstrated by the above embodiments, this specification utilizes a prediction model trained on historical multidimensional data to obtain the predicted redemption rate, redemption rate, and return rate of consumption vouchers, achieving automated integration and quantitative analysis of massive, dynamic data. Based on the predicted values, the expected transaction amount guided by each consumption voucher is determined, and the target consumption voucher set is automatically determined from the candidate set with the objective of maximizing the total global transaction amount. This transforms complex rules and resource constraints into a solvable optimization problem, enabling rapid optimization of the target consumption voucher set. By converting the selected consumption vouchers into thresholds and denominations that conform to the local currency of the target country, the consistency and usability of the operational strategy in a global multi-currency environment are ensured. Accordingly, the embodiments in this specification improve decision-making efficiency, strategy quality, and operational automation in complex cross-border e-commerce scenarios.

[0051] In some embodiments, the sample includes distribution metrics obtained at a specified time point, which are indicators characterizing the distribution status of historical consumption vouchers at the specified time point; the distribution metrics include data associated with specific time points or special events, and the associated data includes: promotional activity type, promotional activity intensity index, or holiday identifier.

[0052] The implementation metrics in this specification include the number of vouchers issued, the number of vouchers redeemed, the redemption rate, the duration of the distribution, and one or more of the following: promotional activity type, promotional activity intensity index, and holiday identifier. The promotional activity type, promotional activity intensity index, and holiday identifier are data associated with specific time points or events, designed to enable the model to perceive and quantify the systematic impact of the external market environment and specific time events on the distribution process of consumer vouchers.

[0053] In the training samples, this data, associated with specific time points or events, is integrated into the feature information of each specified time point as part of dynamic distribution metrics. In the embodiments of this specification, for a certain historical consumer voucher distribution instance, the distribution status metrics collected at a certain time point include data that directly reflects the distribution progress of the vouchers, such as the number of vouchers issued, the number of vouchers redeemed, the redemption rate, and the distribution duration. At the same time, contextual information such as whether the time point is a holiday, whether there is a promotional activity, and the type and intensity of the activity will also be recorded.

[0054] By using data associated with specific times or events as sample data, predictive models can learn that user behaviors such as voucher redemption and redemption do not occur in isolation, but are significantly driven or inhibited by external factors such as large-scale promotions, seasonal shopping festivals, or public holidays. For example, the model may learn that during large-scale promotions, users' willingness to redeem vouchers of the same denomination generally increases; or that the redemption rate of specific types of vouchers varies during certain holidays. Therefore, when applying predictive models, the inherent attributes of vouchers (threshold, denomination) can be combined with the characteristics of the market environment within a specific time period, thereby outputting predicted redemption rates, predicted redemption rates, and predicted returns that are closer to the real market situation in the future. This enhances the model's adaptability to complex and dynamic real business environments and the reliability of its predictions.

[0055] In some embodiments, the samples used to train the prediction model are data sequences constructed from multiple delivery indicators obtained within a preset time window. The preset time window includes multiple specified time points arranged at equal time intervals. Each delivery indicator is obtained at a specified time point, and the delivery indicators in the data sequence are arranged in the order of the specified time points.

[0056] In the embodiments of this specification, the sample construction is a data sequence constructed from multiple deployment indicators arranged in time sequence, which can be represented as: Where d represents the feature dimension, T represents the length of the preset time window, and x i Let represent the i-th delivery metric in the data sequence X, and t represent the specified time point when the latest delivery metric is obtained.

[0057] This preset time window defines the range of observations that the model needs to learn for a given sample. This preset time window can be divided into multiple specified time points. These time points are arranged according to a rule of equal time intervals, such as a fixed time each day or a fixed date each week, thus forming a uniform time grid.

[0058] At each specified time point, a set of distribution indicators reflecting the distribution status of consumption vouchers at that time point is obtained. Each distribution indicator corresponds to a data snapshot at a specific time point. By arranging and combining the distribution indicators obtained at all specified time points within the preset time window according to the chronological order of their corresponding specified time points, a data sequence is formed.

[0059] This construction method transforms the distribution indicator data from a discrete set of points into an ordered sequence, directly and systematically reflecting the continuous evolution of the distribution status of consumption vouchers over time. When such structured data is used as feature input to predictive models, especially when applied to model architectures capable of processing sequential data, the model can more effectively identify and utilize time-dependent dependencies, trends, or periodic patterns.

[0060] Based on the foregoing embodiments, the aforementioned prediction model can be specifically constructed and applied using a Transformer-based neural network architecture. Figure 3 This is a schematic diagram illustrating the training principle of a prediction model provided in an exemplary embodiment.

[0061] This neural network architecture includes a Transformer-based encoder and decoder, and can be applied in this specification to implement the aforementioned claim rate prediction model, redemption rate prediction model, and return rate prediction model. During the training phase, for ease of description, the samples of the prediction model can be denoted as... Where T represents the length of the preset time window, These are the multidimensional feature vectors corresponding to the first to the Tth consecutive time steps, and each multidimensional feature vector... This includes the distribution indicators of consumption vouchers at the corresponding time step i. The sample can be divided into two parts: one part serves as the historical data sequence for model observation and learning, and the other part serves as the target data sequence that the model needs to predict. In the data preprocessing stage, the historical data sequence and the target data sequence in the sample can be extracted by setting three positions: start index, label index, and end index.

[0062] The encoder generates a context vector by extracting temporal dependencies from the input data sequence. The encoder's input is the aforementioned historical data sequence. Figure 3 Taking the sample in the example, the starting index can be set to 1 and the label index can be set to (T-1). Then the extracted historical data sequence can be represented as follows: The encoder is used to perform deep encoding on historical data sequences, extract the temporal dependencies, compress them, and output an intermediate representation, denoted as a context vector. The decoder is used to output a predicted claim rate, predicted redemption rate, or predicted return rate based on the context vector. It can be understood that when the prediction model is a claim rate prediction model, the decoder outputs the predicted claim rate; when the prediction model is a redemption rate prediction model, the decoder outputs the predicted redemption rate; and when the prediction model is a return rate prediction model, the decoder outputs the predicted return rate.

[0063] Figure 4 This is a schematic diagram of the encoder structure of a prediction model provided in an exemplary embodiment. The encoder includes an input layer, a position encoding layer, and multiple encoder layers.

[0064] The input layer is a fully connected network used to map the multidimensional feature vectors at each time step in the historical data sequence to a unified model vector dimension.

[0065] The positional encoding layer assigns positional identifiers to each multidimensional feature vector in the historical data sequence. Using sine and cosine functions, the positional encoding layer generates a unique encoded vector with the same dimensions as the model vector for each multidimensional feature vector in the historical data sequence, serving as the positional identifier. This positional identifier is then element-wise added to the corresponding multidimensional feature vector to generate a combined information sequence containing both content and positional information, which is then output to the encoder layer connected to the positional encoding layer.

[0066] Each encoder layer has an identical structure and is connected sequentially. Each encoder layer performs standardized sub-layer operations and transformations. For example, the encoder includes four structurally identical encoder layers. Each encoder layer includes a self-attention sub-layer and a feedforward neural network sub-layer. Both the self-attention and feedforward neural network sub-layers are followed by a residual normalization module. The self-attention sub-layer calculates the correlation weights between multi-dimensional feature vectors corresponding to all time steps in the historical data sequence, enabling the prediction model to capture the complex global dependencies and cross-time step interaction patterns within the combined information sequence, resulting in an output vector. The feedforward neural network sub-layer includes two fully connected layers for further nonlinear processing and feature integration of the combined information sequence output by the aforementioned self-attention sub-layer. The residual normalization module is used to element-wise add the input and output vectors of the connected self-attention or feedforward neural network sub-layers (residual connection), and then adjust the data distribution of the added vector to a stable range (layer normalization). Following the connection order, the first to third encoder layers output their results to the next encoder layer, and the fourth encoder layer outputs its results to the various decoder layers in the decoder. This output is the context vector. .

[0067] Figure 5 This is a schematic diagram of the decoder structure of a prediction model provided in an exemplary embodiment. The decoder includes an input layer, a positional coding layer, multiple decoder layers, and an output layer.

[0068] The input layer and positional encoding layer of the decoder function similarly to those of the encoder. During model training, the input layer receives the aforementioned target data sequence, and... Figure 3 Taking a sample as an example, the label index is (T-1), the termination index is T, and the target data sequence is the portion extracted from the sample. It can be understood that in practical applications, the length of the target data sequence is specifically set according to the length of the time steps required for prediction by the prediction model. The positional encoding layer configures positional identifiers for each multidimensional feature vector in the target data sequence, thereby generating a combined information sequence that simultaneously contains content information and positional information, and outputs it to the decoder layer connected to the positional encoding layer.

[0069] The first vector of the target data sequence is the same as the last vector of the historical data sequence, both being... The target data sequence uses the last vector of the historical data sequence as the initial input state, and uses this to predict the data at subsequent time steps, outputting a vector. The predicted value can be denoted as a vector. During training, vectors The data in the data can be used as ground truth labels for supervised training; the ground truth labels for the claim rate prediction model are vectors. The truth labels of the claim rate and redemption rate prediction model are vectors. The true labels for the write-off rate and return prediction model are vectors. The rate of return.

[0070] Each decoder layer has the same structure and is connected in sequence. Each decoder layer includes a masked self-attention sub-layer, an encoder-decoder attention sub-layer, and a positional feedforward neural network sub-layer. Feedforward neural network sub-layers are provided between the masked self-attention sub-layer, the encoder-decoder attention sub-layer, and the feedforward neural network sub-layer to perform residual connections and layer normalization operations.

[0071] The masked self-attention sublayer is used to ensure, during attention weight calculation, that any position in the sequence can only "see" the information before that position (including itself) when calculating its attention distribution, and cannot access the information after that position.

[0072] This is the encoder-decoder attention sublayer, used to connect the encoder and decoder. The query vector of this sublayer comes from the output of the previous sublayer of the decoder, while the key and value vectors come from the complete context vector H output by the encoder. Through this attention mechanism, when generating the representation of the current time step, the decoder can dynamically focus on and fuse the most relevant parts of the historical context sequence provided by the encoder, thereby achieving conditional generation based on complete historical information.

[0073] The feedforward neural network sublayer has the same structure and function as the encoder. It consists of two fully connected layers and is used to perform further nonlinear transformation and feature integration on the preceding information.

[0074] Finally, the output, processed by multiple stacked decoder layers, is mapped through the output layer. The output layer consists of a linear transformation layer, which maps the high-dimensional vector sequence output from the top layer of the decoder to the dimension of the target predicted value. For the regression prediction task in this scheme, this output layer directly generates a sequence of predicted values ​​for the target variable corresponding to one or more future time steps, such as the predicted values ​​for the redemption rate, claim rate, or rate of return of consumption vouchers.

[0075] Based on the training principles of the above prediction model, to ensure the model can fully learn and adapt to the dynamic changes in the effectiveness of consumer voucher distribution under different time periods and market promotion rhythms during the model training process, a sliding window mechanism can be introduced to construct training samples. This mechanism continuously scrolls across complete historical time-series data with a fixed-length time window period. Each scroll extracts a corresponding historical data sequence and target data sequence based on the preset historical data size and prediction range, thus forming a new training sample. In this way, a continuous long time-series data can be transformed into a large number of samples that partially overlap in time but have different observation starting points.

[0076] Through the above embodiments, the prediction model can process multi-dimensional and structured time-series delivery indicators, and can also effectively integrate event information at specific time points, thereby achieving a more accurate and reliable prediction of the consumption voucher redemption rate.

[0077] In some embodiments, the target set of consumption vouchers is obtained by solving a first optimization model. The first optimization model has the objective function of maximizing the total global transaction amount and satisfies a first constraint condition, which includes: The number of selected vouchers does not exceed a preset threshold. The sum of the base face value of each selected consumption voucher shall not exceed the preset total budget threshold.

[0078] This embodiment transforms the selection decision for vouchers in the target voucher set into a model-solving problem. The objective function of the first optimization model is set to maximize the total global transaction amount. The total global transaction amount refers to the sum of the expected transaction amounts of all vouchers ultimately selected for the target voucher set. Therefore, the model-solving process is essentially about finding the voucher combination that maximizes this sum.

[0079] To achieve this goal, the model needs to be solved under predefined first constraints. These first constraints characterize the limitations and rules in actual operation. Specifically, one constraint limits the number of selected vouchers, requiring that their total number not exceed a preset threshold. This constraint ensures that the generated voucher scheme is controllable and easily managed in terms of quantity, avoiding difficulties in model solving, user confusion, or excessive operational complexity caused by too many voucher tiers.

[0080] Another constraint in the first set of conditions is to control the total cost of the operation, requiring that the sum of the base denominations of all selected vouchers not exceed a preset total budget threshold. This constraint limits the platform's subsidy costs to a predetermined budget, ensuring that the optimization process must seek to maximize transaction volume within a cost control framework, thereby guaranteeing the feasibility of the operation.

[0081] In the embodiments of this specification, by establishing a first optimization model with the goal of maximizing the total global transaction amount and restricted by the quantity of consumption vouchers and the total budget at the same time, the complex decision-making problem is transformed into a model-solving task. Solving this first optimization model can automatically and efficiently output an optimal or approximately optimal consumption voucher combination plan, that is, the target consumption voucher set, under the given rules and resource constraints.

[0082] In practical applications, the overall average discount rate of the consumption voucher activity is a key indicator to measure the overall profit margin of the platform and the attractiveness of the operation activity. If the average discount rate is too low, it may not be able to effectively stimulate consumption due to insufficient preferential intensity; if the average discount rate is too high, it may lead to an imbalance between the operating cost-benefit ratio and even affect the platform's revenue. Therefore, presetting a reasonable discount rate range can balance the promotion effect and cost control.

[0083] In view of this, further, the first constraint condition also includes that the average value of the discount rates of each selected consumption voucher is within the preset discount rate range. The embodiments of this specification introduce a constraint on the overall preferential intensity of the target consumption voucher set. This constraint condition requires that for all consumption vouchers selected and included in the target consumption voucher set, the average value of the discount rates is calculated based on the discount rates of each consumption voucher, and this average value needs to fall within a certain preset discount rate range. Among them, the discount rate of a single consumption voucher is defined as the ratio of the benchmark denomination of the consumption voucher to the benchmark threshold amount.

[0084] Converting the requirement of the average discount rate into the above constraint condition and incorporating it into the first optimization model can guide the solution process of the model, so that while maximizing the total global transaction amount, the optimization algorithm needs to control the overall average discount rate of the final plan within the specified interval to prevent the optimization result from偏向于过度依赖极高折扣或极低折扣的极端券型组合,使得生成的消费券集合在保持强大成交拉动能力的同时,其整体的优惠水平也符合平台全局的运营策略与成本预期。

[0085] In the target consumption voucher set, if the thresholds of the selected consumption vouchers are concentrated in a narrow range, then a large number of users with consumption capabilities in other intervals will not be able to find consumption vouchers suitable for their order amounts, resulting in the consumption vouchers being unable to effectively reach and convert these users, thus limiting the potential for improving the overall effect of the operation activity.

[0086] In view of this, further, the first constraint condition also includes that for any one of the pre-determined multiple transaction amount intervals, at least one of the selected consumption vouchers has a benchmark threshold amount within the transaction amount interval, and the transaction amount interval is determined based on the historical transaction amount distribution of the target country obtained in advance.

[0087] It should be noted that there is an incorrect expression in the original text of item , "偏向于过度依赖极高折扣或极低折扣的极端券型组合" is a Chinese expression that needs to be corrected to an accurate English description. The corrected translation is as follows: Converting the requirement of the average discount rate into the above constraint condition and incorporating it into the first optimization model can guide the solution process of the model, so that while maximizing the total global transaction amount, the optimization algorithm needs to control the overall average discount rate of the final plan within the specified interval to prevent the optimization result from being biased towards extreme coupon combinations that overly rely on extremely high discounts or extremely low discounts, enabling the generated consumption voucher set to maintain a strong ability to drive transactions while its overall preferential level also conforms to the platform's overall operation strategy and cost expectations.The purpose of this constraint is to ensure that the final set of target consumption vouchers covers user groups with different spending power levels in the target country. The implementation of this constraint relies on the analysis of historical consumption behavior of users in the target country. First, based on the pre-acquired historical transaction amount distribution of the target country, several representative transaction amount intervals are defined. For example, based on the percentiles of historical transaction amounts (such as 25%, 50%, 75%), it can be divided into multiple consecutive intervals of low, medium, and high amounts, each interval corresponding to a group of users with similar spending power. The historical transaction amount distribution can be calculated from pre-acquired data such as the list of products corresponding to each historical consumption voucher, the average order value, and the number of users who made a purchase.

[0088] Based on multiple pre-defined transaction amount ranges, this constraint requires that when solving for the final selected set of target vouchers using the first optimization model, for each of the pre-defined transaction amount ranges, there must be at least one selected voucher whose benchmark threshold amount falls within that range. This constraint guides the search direction of the optimization algorithm, ensuring that while maximizing the total global transaction amount, the first optimization model guarantees sufficient width and coverage in the distribution of threshold amounts for the generated voucher combinations. This prevents the results from being overly biased towards serving only a single consumer level, ensuring that the target voucher set covers a broad user base.

[0089] If multiple coupons with very similar threshold amounts exist within a target set, users may become confused by these similar thresholds, making it difficult for them to intuitively perceive the value differences between different coupon types, thus reducing their willingness to claim and use them. Furthermore, from a platform operation perspective, this can lead to internal competition for resources (such as budget and exposure placement) among overlapping target user groups, resulting in wasted resources and an inability to effectively cover diverse consumption scenarios.

[0090] In view of this, the first constraint further includes that, for any two vouchers in the candidate set, if the difference between the base threshold amounts of the two vouchers is less than a preset difference threshold, at most one of the vouchers will be selected as the voucher.

[0091] This constraint, introduced when screening vouchers from the candidate set, prevents the simultaneous selection of vouchers with excessively close benchmark thresholds. A preset difference threshold defines the quantitative standard for "excessively close." During the optimization decision-making process, any two vouchers in the candidate set are compared. If the absolute difference between the benchmark thresholds of two vouchers is found to be less than the preset difference threshold, then at most one of these vouchers can be selected for the final target voucher set.

[0092] By setting this constraint, the first optimization model needs to select a combination of vouchers that has sufficient differentiation in terms of threshold amount. This ensures that the threshold amount of each type of voucher in the final generated voucher set can show a significant gradient distribution, thereby corresponding to different levels of order amount or spending power, improving the overall coverage efficiency and user experience of the combination, and prompting the budget to be allocated to more differentiated consumption scenarios.

[0093] Within a target set of vouchers, there's a general expectation of a positive correlation between the threshold amount and the face value; users implicitly assume they need to reach a higher spending threshold to receive greater discounts. If a voucher has a lower threshold but a higher face value than another with a higher threshold, it will reduce user trust and willingness to participate in the promotion, contradicting the original intention of vouchers to incentivize users to increase average order value. This leads to a misallocation of service resources and fails to effectively guide users to upgrade their spending levels.

[0094] In view of this, the first constraint further includes that, for any two selected vouchers, if the base threshold amount of the first voucher is lower than the base threshold amount of the second voucher, then the base face value of the first voucher is lower than the base face value of the second voucher.

[0095] In the final selected set of target vouchers, the threshold amount and face value of all vouchers must satisfy the monotonic consistency rule. During the model solution process, for any two different vouchers in the set, they can be denoted as the first consumption right and the second consumption voucher, respectively. If the benchmark threshold amount of the first consumption voucher is confirmed to be lower than that of the second consumption voucher, then the benchmark face value of the first consumption voucher must also be lower than that of the second consumption voucher. That is, within the entire selected voucher combination, the benchmark face value must increase with the increase of the benchmark threshold amount, thereby ensuring the logical rationality of the operation activities and enabling the vouchers to more effectively guide user consumption behavior and increase the platform's average order value and transaction volume.

[0096] Based on the foregoing embodiments, an application example of the first optimization model in this specification can be represented as follows: , , , , , , in, Let represent the decision variable, indicating whether the i-th voucher in the candidate set is selected. Q represents the total number of vouchers in the candidate set. Si Let P be the base face value of the i-th consumption voucher. i Let c be the base threshold amount for the i-th consumption voucher. i Let l be the redemption rate of the i-th consumption voucher. i Let r be the redemption rate of the i-th consumption voucher. i Let i be the rate of return for the i-th consumption voucher. Let N be the expected transaction amount driven by the i-th consumption voucher, which is the product of the redemption rate, the benchmark threshold amount, the redemption rate, and the return rate, and let N be the number of exposed users. This represents the total budget threshold. z is the standard average discount rate. The aforementioned average discount rate range, is the preset discount rate deviation constant. K is the quantity threshold in the target set of consumer vouchers.

[0097] M is a preset constant, usually defined as a large positive integer value, such as 100,000. This specification does not impose this restriction. This constraint, based on the defined positive integer value M, ensures... and In every possible value case, it must be satisfied that if the threshold amount P of consumption voucher i is... i The amount P below the threshold of consumption voucher j j Then the face value S of consumption voucher i i It is also lower than the face value of the consumption voucher J. j .

[0098] a and b are preset range control parameters. For example, a takes the value 0.8 and b takes the value 1.2. This is the reference value for the m-th transaction amount interval determined based on the historical transaction amount distribution. The corresponding constraints ensure that within each preset transaction amount interval... Within this scope, at least one selected voucher must have a minimum spending threshold. This will allow us to reach users with different spending power levels. The transaction amount interval is determined based on the historical transaction amount distribution of the target country obtained in advance, or it is denoted as the quantile number. For example, if the historical transaction amount distribution is divided into six transaction amount intervals, then m=1,2,...,6.

[0099] In this application example, the items in the first constraint condition can mutually restrict and cooperate with each other during the optimization process, jointly guiding the solution direction.

[0100] For example, the quantity limit constraint prevents the combination from becoming overly complex, while the coverage range constraint requires that this limited number of vouchers be distributed across different spending levels. The combination of the quantity constraint and the constraint of covering different spending ranges ensures that the solution achieves user coverage with a streamlined tier structure while avoiding the ineffective accumulation of resources in a single range.

[0101] For example, the threshold constraint for difference in minimum spending amount and the consistency constraint for minimum spending amount together enhance the logical consistency of voucher selection. The difference threshold constraint avoids user confusion and competition caused by thresholds being too close, forming a clear amount gradient; while the consistency constraint ensures that the discount increases monotonically with the spending requirement at this gradient, which aligns with users' basic understanding. These two constraints make the generated voucher combinations not only clearly graded but also logically consistent, allowing users to quickly understand and make decisions.

[0102] All the above constraints work in conjunction with the core objective function of maximizing the total global transaction amount. The solution process seeks the globally optimal solution while satisfying a series of constraints, including coverage breadth (interval constraints), structural clarity (difference thresholds, consistency constraints), and cost controllability (quantity, budget, and discount rate constraints). The synergy of these constraints ensures that the final set of target consumer vouchers is not an extreme case in a single indicator, but rather an optimal or near-optimal solution that performs well across multiple key dimensions. It possesses a reasonable cost structure, meets clear user perception, and achieves broad customer coverage, thereby comprehensively ensuring operational efficiency, risk control, and user experience.

[0103] It should be noted that the selection of various constraints in this specification is based on a comprehensive consideration of "operational resources" and "reasonable tiering." Regarding operational resources, this may include an upper limit on the number of vouchers and a total budget threshold, the purpose of which is to ensure the operational feasibility and cost control of the generated scheme. Regarding reasonable tiering, this includes the average discount rate range, threshold difference threshold, transaction amount range, and threshold face value consistency requirements, the purpose of which is to maintain the logical consistency and tiered rationality of the values ​​within the voucher combination, and to ensure that users at different spending power levels are effectively covered.

[0104] Experimental data demonstrates the crucial importance of this carefully designed constraint system. Excessively reducing constraints, such as weakening the guarantee of transaction structure (e.g., removing constraints on transaction amount ranges), will lead the optimization process to favor only high-denomination, high-threshold vouchers that generate the highest "unit budget transaction amount," resulting in a severely biased solution towards users with higher transaction amounts. This harms the experience for users with lower to medium transaction amounts and hinders the potential of voucher distribution to increase overall transaction volume. Conversely, adding too many or overly strict constraints without considering operational realities can easily lead to an infeasible optimization model or an overly conservative solution that fails to fully leverage the incentive effect of the budget, thus failing to achieve the operational goal of approaching the optimal solution. Therefore, the current constraint set was developed based on thorough data analysis, historical effect attribution, and multiple simulation experiments to generate a voucher combination that simultaneously possesses high transaction-driving capacity, reasonable cost-effectiveness, and a good user coverage experience.

[0105] The conversion between the threshold amount and the face value for the target currency in a target country can be achieved through real-time exchange rate conversion. However, directly using exchange rate multiplication often results in non-integer amounts or values ​​that do not conform to the common payment habits of users in the target market (e.g., the face value or threshold has an unusual decimal). Such amounts can negatively impact user experience and the willingness to use vouchers in real-world business scenarios.

[0106] In one or more embodiments of this specification, the conversion to the threshold amount and denomination for the target currency corresponding to the target country is achieved by solving a second optimization model. The second optimization model takes minimizing the conversion deviation index as the objective function and satisfies the second constraint condition. The conversion deviation is determined based on the deviation of the benchmark threshold amount before and after the conversion and the deviation of the benchmark denomination before and after the conversion. The second constraint includes: The difference in discount rates before and after the conversion is within the preset discount rate deviation range.

[0107] The goal of the second optimization model is to calculate an optimal set of target currency threshold amounts and face values ​​for each selected benchmark consumption voucher, while meeting the requirements of localized services.

[0108] The objective function of the second optimization model is set to minimize a conversion deviation index. This conversion deviation index comprehensively measures the degree of deviation between the conversion result and the original benchmark value. Specifically, it is determined based on the deviation between the threshold amount of the converted target currency and the theoretical threshold amount calculated from the benchmark threshold amount exchange rate, as well as the deviation between the face value of the converted target currency and the theoretical face value calculated from the benchmark face value exchange rate.

[0109] An application example of the second optimization model can be represented as follows: , , , , in, Let be the denomination of the target currency to be solved. Let represent the threshold amount in the target currency to be solved. The base currency represents the base denomination. The threshold amount is expressed in the base currency. r is the exchange rate from the base currency to the target currency. This sets the maximum allowable discount rate deviation.

[0110] Solving the second optimization model requires satisfying a pre-defined second constraint. One constraint is that the discount rate of the vouchers before and after the conversion must remain basically stable. Specifically, the deviation between the discount rate of the converted vouchers (i.e., the ratio of the target face value to the target threshold amount) and the discount rate of the base vouchers before the conversion (i.e., the ratio of the base face value to the base threshold amount) must be controlled within a pre-defined, small range of discount rate deviation. This constraint ensures that the discount of the vouchers is maintained after currency conversion, avoiding significant differences in discount rates across countries due to the conversion process, thus guaranteeing consistent discounts across different national markets for global operations.

[0111] To better align with the widely accepted digital preferences and payment habits in the target country's local market, the second constraint further includes that the converted threshold amount and denomination conform to a pre-defined pattern for the target country. The pre-defined pattern includes at least one of the last digit of the constraint threshold amount and the non-highest digit of the constraint denomination.

[0112] This constraint requires that the converted threshold amount and denomination follow a specific pattern pre-defined for the target country. For example, the threshold amount and denomination can be constrained to be integers.

[0113] For example, the last digit of the threshold amount can also be constrained. In many consumer markets, the last digit of the price or discount threshold amount is often customarily set to a specific number, such as 9, 8, 5, or 0, to align with local consumers' price perceptions and enhance the attractiveness of the voucher.

[0114] For example, constraints can also be imposed on the non-highest digits of the denomination. This refers to restricting the digits in the denomination other than the digits representing the largest place value (such as the hundreds and tens digits). For instance, it might be required that the tens and units digits of the denomination avoid small numbers like 1 or 2, or tend to form "multiple tens" or "half-hundreds" structures like 50, to make the discount amount appear more regular and easier to calculate.

[0115] The embodiments in this specification can automatically generate target currency amounts that are as close as possible to the exchange rate conversion in terms of numerical value, while maintaining the discount level in substance, and are integers. This effectively overcomes the problems of non-integer amounts or numerical values ​​that may be generated by simple conversion, achieving currency localization while maintaining the consistency of operational strategies and the rationality of user experience.

[0116] In some embodiments, the method further includes displaying the predicted claim rate, predicted redemption rate, and predicted return rate in a visualized interface. This visualized interface, in specific applications, can be an operations management interface used by operations personnel. This interface can directly present the predicted claim rate, predicted redemption rate, and predicted return rate to operations personnel. The presentation can be the numerical values ​​of the predicted claim rate, predicted redemption rate, and predicted return rate at several future time points currently output by each prediction model; or it can be a chart, table, or list generated by combining the actual claim rate, redemption rate, and return rate collected at several historical time points with the predicted claim rate, predicted redemption rate, and predicted return rate at several future time points. Displaying the prediction results in a visualized interface allows operations personnel to see the possible distribution results before distributing selected consumer vouchers, thus providing a quantitative basis for final decision-making. Operations personnel can use this to confirm, fine-tune, or even reject the distribution strategy composed of each selected consumer voucher.

[0117] Based on the foregoing embodiments, the method for generating cross-border e-commerce consumer vouchers provided in this specification can be deployed in the system architecture of a cross-border e-commerce platform as a software module.

[0118] Figure 6 This is a schematic diagram illustrating an application example of a method for generating cross-border e-commerce vouchers, provided by an exemplary embodiment. In this application example, the method for generating cross-border e-commerce vouchers is implemented through an operation service module within the cross-border e-commerce platform. This operation service module collaborates with other existing service modules within the platform and external data services through predefined interfaces and data flows to achieve a closed loop from historical data analysis and voucher generation to strategy feedback.

[0119] For example, the architecture of this cross-border e-commerce platform can be logically divided into a data support layer, an intelligent service layer, and an execution management layer. The data support layer, based on the platform's data warehouse, is used to aggregate and store historical data from various execution management layer modules. The execution management layer includes order management, user management, and activity management modules. These modules directly handle the platform's daily transactions, promotions, user management, and activity operations, and store process data in their associated databases. The databases associated with the order management, user management, and activity management modules are respectively denoted as the order management database, user management database, and activity management database. The data support layer extracts, transforms, and loads data from these databases into the data warehouse through periodic data integration tasks, forming a cross-layered integrated data view.

[0120] The intelligent service layer is the layer where the operations service module resides. The operations service module relies on integrated historical data provided by the data support layer for strategy calculations. It queries the data warehouse to obtain the structured historical dataset needed to train predictive models and execute optimization decisions. When the calculation process involves currency conversion, this module calls an external currency data service interface to obtain real-time information. After completing the calculations, the operations service module distributes the generated voucher plan to the activity management module of the execution management layer via the inter-service communication interface.

[0121] After receiving the plan, the activity management module of the executive management team is responsible for the specific creation, rule configuration, and formal distribution of consumption vouchers to users within the platform. Users interact with these consumption vouchers through the platform's front-end application. Their actions such as claiming, placing orders, and redeeming vouchers trigger the corresponding functions of the activity management module and the order service module. These action results are stored as new data in their respective databases.

[0122] This newly generated data is then synchronized from the database to the data warehouse through the integration process of the data support layer, forming the latest performance data record. In subsequent decision-making cycles, the operations service module can retrieve this updated data from the data warehouse to evaluate the effectiveness of historical strategies and iterative model parameters, thereby achieving an automated closed loop in the strategy optimization process.

[0123] Figure 7 This is a schematic structural diagram of a device provided in an exemplary embodiment. For example... Figure 7As shown, device 700 mainly consists of a communication interface 702, a user interface 704, a processor 706, and a data storage 708. These components are interconnected and communicate with each other via a system bus, network, or other connection mechanism 710. The communication interface 702 enables device 700 to communicate with other devices, access networks, and transmission networks via analog or digital modulation. For example, the communication interface 702 may include a chipset and antenna for wireless communication with a radio access network or access point. Furthermore, the communication interface 702 can be a wired interface such as Ethernet, Token Ring, or a USB port, or a wireless interface such as Wi-Fi, Bluetooth, Global Positioning System (GPS), or a wide-area wireless interface (e.g., WiMAX or LTE). Of course, the communication interface 702 can also support other forms of physical layer interfaces and standard or proprietary communication protocols. The communication interface 702 may also include multiple physical communication interfaces, such as Wi-Fi, Bluetooth, and wide-area wireless interfaces.

[0124] User interface 704 includes receiving user input and providing output to the user. Therefore, user interface 704 may include input components such as a keypad, keyboard, touch-sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera, and video camera, and output components such as a display screen (which may be combined with a touch-sensitive panel), CRT, LCD, LED, display using DLP technology, printer, and other similar devices known or developed in the future. User interface 704 may also generate auditory output via speakers, speaker jacks, audio output ports, audio output devices, headphones, and other similar devices known or developed in the future. In some embodiments, user interface 704 may include software, circuitry, or other forms of logic capable of transmitting and receiving data from external user input / output devices. Additionally or alternatively, device 700 may support remote access from other devices via communication interface 702 or another physical interface (not shown). User interface 704 may be configured to receive user input, the position and movement of which may be indicated by indicators or cursors described herein. User interface 704 may also be configured as a display device for rendering or displaying text fragments.

[0125] Processor 706 may contain one or more general-purpose processors and / or special-purpose processors.

[0126] Data storage 708 may include one or more volatile and / or non-volatile storage components and may be integrated wholly or partially with processor 706. Data storage 708 may include removable and non-removable components.

[0127] Processor 706 is capable of executing program instructions 718 (e.g., compiled or uncompiled program logic and / or machine code) stored in data storage 708 to perform the various functions described herein. Data storage 708 may contain a non-transitory computer-readable medium on which program instructions are stored, which, when executed by device 700, enable device 700 to perform any methods, processes, or functions disclosed in this specification and / or the accompanying drawings. Processor 706 executing program instructions 718 may result in processor 706 using data 712.

[0128] For example, program instructions 718 may include an operating system 722 (e.g., an operating system kernel, device drivers, and / or other modules) installed on device 700 and one or more applications 720 (e.g., a browser, social application, or game application). Similarly, data 712 may include operating system data 716 and application data 714. Operating system data 716 is primarily accessible to the operating system 722, while application data 714 is primarily accessible to one or more applications 720. Application data 714 may reside in a file system visible or hidden from the user of device 700.

[0129] Application 720 can communicate with operating system 722 through one or more application programming interfaces (APIs). These APIs help application 720 read and / or write application data 714, transmit or receive information via communication interface 702, receive or display information on user interface 704, etc.

[0130] In some terminology, application 720 may be simply referred to as "app". Furthermore, application 720 can be downloaded to device 700 through one or more online app stores or app markets. However, applications can also be installed on device 700 in other ways, such as through a web browser or a physical interface on device 700 (e.g., a USB port).

[0131] Figure 8 This is a schematic structural diagram of a cross-border e-commerce consumer voucher generation system provided in an exemplary embodiment. The system includes: The sample construction unit 801 is used to construct three sample sets with ground truth labels based on the distribution indicators. The distribution indicators are metrics that characterize the distribution status of historical consumption vouchers at a specified point in time. The samples in the first sample set use the historical redemption rate as the ground truth label, the samples in the second sample set use the historical redemption rate as the ground truth label, and the samples in the third sample set use the historical return rate as the ground truth label. The sample construction unit uses the distribution indicators of historical consumption vouchers as input, combined with the static attributes of the historical consumption vouchers (threshold amount, face value, and distribution country), to construct three types of datasets, which are then output to the model training unit to provide the data foundation required for training the redemption rate prediction model, the redemption rate prediction model, and the return rate prediction model, respectively.

[0132] Model training unit 802 is used to train a claim rate prediction model based on a first sample set, a redemption rate prediction model based on a second sample set, and a return rate prediction model based on a third sample set. The execution unit 803 is used to determine a candidate set of consumption vouchers corresponding to a selected base currency. Each consumption voucher has a corresponding base threshold amount and base face value. Based on the selected target country, for each consumption voucher in the candidate set, the predicted redemption rate, predicted redemption rate, and predicted return rate are obtained through a redemption rate prediction model, a redemption rate prediction model, and a return rate prediction model, respectively. Based on the predicted redemption rate, predicted redemption rate, and predicted return rate of each consumption voucher, the expected transaction amount to be guided by each consumption voucher is determined. Based on the expected transaction amount guided by each consumption voucher, with the objective of maximizing the total global transaction amount, a target consumption voucher set for the target country is determined from the candidate set. The target consumption voucher set includes at least one consumption voucher selected from the candidate set, and the total global transaction amount is the sum of the expected transaction amounts guided by each selected consumption voucher. The base threshold amount and base face value corresponding to each selected consumption voucher are converted into the threshold amount and face value of the target currency corresponding to the target country, and the selected consumption vouchers are then distributed. The effect feedback unit 804 is used to collect distribution metrics and send them to the sample construction unit after the selected consumer vouchers are distributed. To enable the model to learn in a timely manner about the distribution effects of newly distributed consumer vouchers during actual operation, the effect feedback unit can automatically collect the latest distribution metrics related to each selected consumer voucher and send them back to the sample construction unit. In this way, the distribution metrics obtained under the influence of each new distribution activity can be automatically collected and become part of the sample data used in the next round of model training. Furthermore, it can be set to drive the sample construction unit to automatically build a new sample set after accumulating a sufficient number of newly collected distribution metrics, thereby driving the model training unit to perform periodic or triggered retraining of the prediction model. This allows the prediction model to continuously absorb the latest market feedback and changes in user behavior, thereby maintaining or even improving its predictive accuracy and adaptability.

[0133] Figure 9 This is a schematic structural diagram of an exemplary embodiment of a cross-border e-commerce voucher generation device. This cross-border e-commerce voucher generation device can be applied to, for example... Figure 7 In the device shown, or applied to Figure 8 The system shown is an execution and distribution unit to implement the technical solution described in this specification. The device for generating cross-border e-commerce consumer vouchers may include: The candidate acquisition module 901 is used to determine the candidate set of consumption vouchers corresponding to the selected base currency, wherein the consumption vouchers have corresponding base threshold amount and base face value. The indicator prediction module 902 is used to obtain the predicted redemption rate, predicted redemption rate, and predicted return rate of each consumption voucher in the candidate set based on the selected target country, using pre-trained redemption rate prediction models, redemption rate prediction models, and return rate prediction models, respectively. The redemption rate prediction model, the redemption rate prediction model, or the return rate prediction model is trained using multiple samples with ground truth labels. Each sample includes the threshold amount, face value, distribution country, and distribution indicator obtained at a specified time point for a historical consumption voucher. The distribution indicator is an indicator representing the distribution status of the historical consumption voucher at the specified time point. The ground truth label for the sample of the redemption rate prediction model is the historical redemption rate, the ground truth label for the sample of the redemption rate prediction model is the historical redemption rate, and the ground truth label for the sample of the return rate prediction model is the historical return rate. It is also used to determine the expected transaction amount guided by each of the aforementioned consumption vouchers based on the predicted redemption rate, predicted redemption rate, and predicted return rate of each of the aforementioned consumption vouchers. The operations optimization module 903 is used to determine the target consumer voucher set for the target country from the candidate set based on the expected transaction amount of each of the consumer vouchers, with the goal of maximizing the total global transaction amount. The target consumer voucher set includes at least one consumer voucher selected from the candidate set, and the total global transaction amount is the sum of the expected transaction amounts of each of the selected consumer vouchers. The currency conversion module 904 is used to convert the base threshold amount and base face value corresponding to each of the selected consumption vouchers into the threshold amount and face value of the target currency corresponding to the target country.

[0134] In one or more embodiments, the target set of consumption vouchers is obtained by solving a first optimization model, the first optimization model having the objective function of maximizing the total global transaction amount and satisfying a first constraint condition, the first constraint condition including: The number of selected vouchers does not exceed a preset threshold. The sum of the base face value of each selected consumption voucher shall not exceed the preset total budget threshold.

[0135] In one or more embodiments, the first constraint further includes that the average discount rate of each selected voucher is within a preset discount rate range.

[0136] In one or more embodiments, the first constraint further includes that, for any one of a plurality of predetermined transaction amount ranges, at least one of the selected vouchers has a benchmark threshold amount within the transaction amount range, the transaction amount range being determined based on the historical transaction amount distribution of the target country obtained in advance.

[0137] In one or more embodiments, the first constraint further includes that, for any two vouchers in the candidate set, if the difference between the base threshold amounts of the two vouchers is less than a preset difference threshold, at most one of the vouchers will be selected as the voucher.

[0138] In one or more embodiments, the first constraint further includes that, for any two selected vouchers, if the base threshold amount of the first voucher is lower than the base threshold amount of the second voucher, then the base face value of the first voucher is lower than the base face value of the second voucher.

[0139] In one or more embodiments, the delivery metrics include data associated with specific time points or events, such as promotional activity type, promotional activity intensity index, or holiday identifier.

[0140] In one or more embodiments, the sample is a data sequence constructed by multiple delivery indicators obtained within a preset time window. The preset time window includes multiple specified time points arranged at equal time intervals. Each delivery indicator is obtained at one of the specified time points, and the delivery indicators in the data sequence are arranged in the order of the specified time points.

[0141] In one or more embodiments, the conversion to a threshold amount and denomination for the target currency corresponding to the target country is achieved by solving a second optimization model. The second optimization model takes minimizing the conversion deviation index as the objective function and satisfies a second constraint condition. The conversion deviation is determined based on the deviation of the benchmark threshold amount before and after the conversion and the deviation of the benchmark denomination before and after the conversion. The second constraint includes: The difference in discount rates before and after the conversion is within the preset discount rate deviation range.

[0142] In one or more embodiments, the second constraint further includes that the converted threshold amount and denomination conform to a predetermined pattern for the target country, the predetermined pattern including at least one of constraining the last digit of the threshold amount and constraining the non-highest digit of the denomination.

[0143] For ease of description, the above devices are described by dividing them into various modules or units based on their functions. Of course, when implementing one or more of these specifications, the functions of each module or unit can be implemented in the same or different software and / or hardware, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0144] Based on the same concept as the methods described above, this specification also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor performs the steps of the method as described in any of the above embodiments by executing the executable instructions.

[0145] Based on the same concept as the methods described above, this specification also provides a computer-readable storage medium having computer instructions stored thereon that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0146] Based on the same concept as the methods described above, this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0147] What those skilled in the art will understand is: In this specification, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitation, the presence of additional identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded.

[0148] In this specification, “a,” “an,” and “the” do not specifically refer to the singular, but may also include the plural.

[0149] In this specification, ordinal numbers such as "first," "second," etc., do not necessarily indicate order; they are often used to distinguish between objects. For example, "first server" and "second server" usually refer to two servers. To differentiate between these two servers, they are described as "first server" and "second server." Of course, sometimes these two servers may be the same server.

[0150] In this specification, unless explicitly stated otherwise, "receiving and sending data" does not necessarily mean direct receiving and sending; it can also mean indirect receiving and sending. For example, A receiving data sent by B can be understood as A directly receiving the data sent by B, or it can be understood as A indirectly receiving the data sent by B through other entities such as C. Similarly, B sending data to A can be understood as B sending the data directly to A, or it can be understood as B indirectly sending the data to A through other entities such as C. Here, C can be one entity, or it can be two or more entities.

[0151] In this specification, unless explicitly stated otherwise, the relationships between structures can be direct or indirect. For example, when describing "A is connected to B," unless it is explicitly stated that A and B are directly connected, it should be understood that A can be directly connected to B or indirectly connected to B. Similarly, when describing "A is on top of B," unless it is explicitly stated that A is directly above B (AB is adjacent and A is above B), it should be understood that A can be directly above B or indirectly above B (AB is separated by other elements, and A is above B). And so on.

[0152] This specification uses specific terms to describe embodiments thereof. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of those different embodiments or examples, without contradiction.

[0153] Although one or more embodiments of this specification provide method steps as described in the embodiments or flowcharts, it is understood that the order of steps listed in the embodiments or flowcharts is only one of many possible execution orders and does not represent the only execution order. Therefore, when the claims involve method steps, any changes or adjustments to the order of such steps, or the parallelism between steps, are also within the scope of protection of the claims.

Claims

1. A method for generating cross-border e-commerce consumer vouchers, comprising: Based on the selected base currency, a candidate set of consumption vouchers corresponding to the base currency is determined, and the consumption vouchers have corresponding base threshold amounts and base denominations; Based on the selected target countries, for each voucher in the candidate set, the predicted redemption rate, predicted redemption rate, and predicted return rate of the voucher are obtained using pre-trained redemption rate prediction models, redemption rate prediction models, and return rate prediction models, respectively. The redemption rate prediction model, the redemption rate prediction model, or the return rate prediction model is trained using multiple samples with ground truth labels. The ground truth labels of the samples in the redemption rate prediction model are historical redemption rates, the ground truth labels of the samples in the redemption rate prediction model are historical redemption rates, and the ground truth labels of the samples in the return rate prediction model are historical return rates. Based on the predicted redemption rate, predicted redemption rate, and predicted return rate of each of the aforementioned consumption vouchers, the expected transaction amount guided by each of the aforementioned consumption vouchers is determined. Based on the expected transaction amount of each of the aforementioned consumption vouchers, with the goal of maximizing the total global transaction amount, a target consumption voucher set for the target country is determined from the candidate set, wherein the target consumption voucher set includes at least one consumption voucher selected from the candidate set, and the total global transaction amount is the sum of the expected transaction amounts of each of the selected consumption vouchers. The base threshold amount and base face value corresponding to each of the selected consumption vouchers are converted into the threshold amount and face value in the target currency corresponding to the target country.

2. The generation method according to claim 1, wherein the target set of consumer vouchers is obtained by solving a first optimization model, the first optimization model having the objective function of maximizing the total global transaction amount and satisfying a first constraint condition, the first constraint condition including: The number of selected vouchers does not exceed a preset threshold. The sum of the base face value of each selected consumption voucher shall not exceed the preset total budget threshold.

3. The generation method according to claim 2, wherein the first constraint further includes that the average discount rate of each selected consumer voucher is within a preset discount rate range.

4. The generation method according to claim 2, wherein the first constraint further includes, for any one of a plurality of predetermined transaction amount ranges, at least one of the selected consumption vouchers has a benchmark threshold amount within the transaction amount range, wherein the transaction amount range is determined based on the historical transaction amount distribution of the target country obtained in advance.

5. The generation method according to claim 2, wherein the first constraint further includes, for any two vouchers in the candidate set, if the difference between the base threshold amounts of the two vouchers is less than a preset difference threshold, at most one of the vouchers shall be selected as the voucher.

6. The generation method according to claim 2, wherein the first constraint further includes, for any two selected vouchers, if the base threshold amount of the first voucher is lower than the base threshold amount of the second voucher, then the base face value of the first voucher is lower than the base face value of the second voucher.

7. The generation method according to claim 1, wherein the sample includes distribution indicators obtained at a specified time point, and the distribution indicators are indicators characterizing the distribution status of historical consumption vouchers at the specified time point; The delivery indicator includes data associated with a special time point or a special event, the associated data including: Promotional activity type, promotional activity intensity index, or holiday identifier.

8. The generation method according to claim 7, wherein the sample is a data sequence constructed based on a plurality of the deployment indicators obtained within a preset time window, the preset time window including a plurality of specified time points arranged at equal time intervals, each deployment indicator being obtained at one of the specified time points, and the deployment indicators in the data sequence being arranged in the order of the specified time points.

9. The generation method according to claim 1, wherein the conversion into a threshold amount and denomination for the target currency corresponding to the target country is achieved by solving a second optimization model, wherein the second optimization model takes minimizing the conversion deviation index as the objective function and satisfies a second constraint condition, and the conversion deviation is determined based on the deviation of the benchmark threshold amount before and after the conversion and the deviation of the benchmark denomination before and after the conversion; The second constraint includes: The difference in discount rates before and after the conversion is within the preset discount rate deviation range.

10. The generation method according to claim 9, wherein the second constraint further includes that the converted threshold amount and denomination conform to a predetermined pattern for the target country, the predetermined pattern including at least one of constraining the last digit of the threshold amount and constraining the non-highest digit of the denomination.

11. The generation method according to claim 1, wherein at least one of the claim rate prediction model, the redemption rate prediction model, and the return rate prediction model employs a Transformer-based neural network architecture, the neural network architecture comprising an encoder and a decoder, the encoder being used to generate a context vector by extracting temporal dependencies in the input data sequence, and the decoder being used to output the predicted claim rate, the predicted redemption rate, or the predicted return rate based on the context vector.

12. The method of generating of claim 1, the method further comprising: The predicted claim rate, predicted redemption rate, and predicted return rate are displayed in a visual interface.

13. A system for generating cross-border e-commerce consumer vouchers, comprising: The sample construction unit is used to construct a first sample set, a second sample set, and a third sample set with truth labels based on the distribution indicators. The distribution indicators are indicators that characterize the distribution status of historical consumption vouchers at a specified time point. The samples in the first sample set use the historical redemption rate as the truth label, the samples in the second sample set use the historical redemption rate as the truth label, and the samples in the third sample set use the historical return rate as the truth label. The model training unit is used to train a claim rate prediction model based on the first sample set, a redemption rate prediction model based on the second sample set, and a return rate prediction model based on the third sample set. The execution unit is used to determine a candidate set of consumption vouchers corresponding to a selected base currency, wherein each consumption voucher has a corresponding base threshold amount and base face value; based on a selected target country, for each consumption voucher in the candidate set, the predicted redemption rate, predicted redemption rate, and predicted return rate of the consumption voucher are obtained through the redemption rate prediction model, the redemption rate prediction model, and the return rate prediction model, respectively; based on the predicted redemption rate, predicted redemption rate, and predicted return rate of each consumption voucher, the expected transaction amount to be guided by each consumption voucher is determined; based on the expected transaction amount to be guided by each consumption voucher, with the objective of maximizing the total global transaction amount, a target consumption voucher set for the target country is determined from the candidate set, wherein the target consumption voucher set includes at least one consumption voucher selected from the candidate set, and the total global transaction amount is the sum of the expected transaction amounts to be guided by each selected consumption voucher; the base threshold amount and base face value corresponding to each selected consumption voucher are converted into a threshold amount and face value for the target currency corresponding to the target country, and the selected consumption vouchers are then distributed; The effect feedback unit is used to collect distribution indicators and send them to the sample construction module after the selected consumer vouchers are distributed.

14. An electronic device comprising: processor; A memory for storing processor-executable instructions; wherein the processor implements the steps of the method as described in any one of claims 1-12 by executing the executable instructions.

15. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1-12.

16. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1-12.

Images